EIP-7707 Incentivizing access list provision (Incentivize Access List Provisioning)
No call decision · Fees & gas · EIP status Stagnant · created 2024-05-12
A proposal to significantly lower the gas cost of including an access list in a transaction, to encourage users to actually use access lists to speed up block execution.
Why it is needed
The access list introduced by EIP-2930 is a mechanism for declaring in advance which addresses and storage slots a transaction will touch, but its cost structure means it is almost never used. Including a single address requires 24 storage keys to break even, and each unused key included but not actually used costs 1900 gas, so even a single unused key means 43 keys are needed to break even. Meanwhile, some clients spend more than 70% of block execution time on data loading, because sequential transaction execution accesses storage in what is effectively a random pattern. Even fast storage like NVMe drives only reach peak throughput when multiple requests are issued concurrently, and if an access list is provided in advance, the client can know which data to load first and hide this latency.
What changes
Today
- Charges 1900 gas per storage key and 2400 gas per address
- Needs 24 storage keys per address just to break even
- A single unused key means 43 keys are needed to avoid a net loss
- In practice almost no transactions include access lists
After EIP-7707
- Lowers the storage key cost from 1900 to 320 gas and the address cost from 2400 to 512 gas
- Brings the price in line with calldata
- For the same number of keys, including an access list becomes advantageous
- Makes it easier for clients to know in advance which data to prefetch before execution
Who is affected
- Users: Including an access list in a transaction now carries less risk of loss and yields a bigger gas saving.
- Wallet & app developers: The eth_createAccessList function for attaching a generated list to a transaction becomes practically useful.
- Client teams: Prefetching data based on the access list creates room to reduce storage access latency caused by sequential execution.
In depth
Background of the problem
The access list in EIP-2930 is a mechanism for pre-warming SLOAD to save gas, but because it only pays off once 24 or more storage keys are included, and each unused key costs 1900 gas, it is almost never used in practice. In some clients, more than 70% of block execution time goes to data loading, because sequential transaction execution repeatedly probes storage in what amounts to a random pattern. NVMe drives achieve high throughput through concurrent processing across multiple queues, but lose that performance when requests are made one at a time, sequentially. Network-attached storage suffers even greater latency. If a node knows in advance what data a block execution will need, it can process data access in parallel ahead of time, similar to instruction pipelining in a CPU, hiding this latency. This could also serve as a data dependency hint for future parallel transaction execution.
Constants being changed
The two gas parameters of EIP-2930 are changed as follows: ACCESS_LIST_STORAGE_KEY_COST is lowered from 1900 to 320, and ACCESS_LIST_ADDRESS_COST from 2400 to 512. This is a minimal change that swaps only these two values while leaving all other logic untouched.
Basis for the pricing
The new costs follow the calldata pricing model directly. They use STANDARD_TOKEN_COST * tokens_in_access_lists, where tokens_in_access_lists is computed as bytes_in_access_lists * 4. Since an address is 32 bytes, 32*4*4=512, and since a storage key is 20 bytes, 20*4*4=320. In other words, data placed in an access list is priced the same as sending a small amount of calldata, aligning it with the standard calldata baseline rather than the original 'bandwidth cost recovery' rationale that EIP-2930 was built on.
Relationship to earlier proposals
This is not the first attempt to lower access list costs. EIP-3521 proposed a similar direction but only lowered ACCESS_LIST_ADDRESS_COST. This proposal is broader in scope, lowering both the address cost and the storage key cost.
Example of the break-even shift
According to the table in the original document, for a cold SLOAD including 10 storage keys, the current cost is OP cost 100 + storage keys 1900 + address 240 for a total of 2240 gas, while under the proposal it becomes OP cost 100 + storage keys 320 + address 51.2 for a total of 471 gas. For 50 storage keys, the cost drops from 2048 gas to 430 gas, meaning the relative burden of including an access list shrinks regardless of key count.
Backwards compatibility and security
This EIP is a small modification to EIP-2930, changing only two gas parameters while keeping the rest of the mechanism unchanged. The security considerations are stated to be identical to those of EIP-2930, the position being that this change introduces no new security risk.
Current status
This EIP is in Stagnant status, not currently under active discussion or decided for fork inclusion. The call records and debate materials provided contain no separate decision or discussion on this proposal.